visitor design pattern, loki and more

A

aaragon

Hello everyone,

I've been trying to work with the visitor design pattern, and it works
fine except for the following.
Let's suppose that we have a fixed hierarchy of classes (many of them)
which I cannot modify. I decided to use the visitor design pattern
depending on the actual type of the classes because those classes
already support the loki visitor.

#include <Loki/Visitor.h>

// forward declarations
class Class_base;
class Class1;
class Class2;
....
class Class20;

// Class list, all derive from Class_base and they all accept the
visitor design pattern through
// the loki visitor
typedef LOKI_TYPELIST_20(Class1, Class2,...,Class20) ClassList;
//! Class visitor
typedef Loki::CyclicVisitor<void, ClassList > ClassVisitor;


class VisitorImpl : public ClassVisitor {

// treat visitor as a functor
void operator(Class_base& b) { b.Accept(*this); }

ReturnType Visit(Class1& c1) { // do stuff with Class1; }
ReturnType Visit(Class2& c2) { // do stuff with Class2; }
...
ReturnType Visit(Class20& c20) { // do stuff with Class20; }
};


Well, this works well, but it seems very silly to add 20 different
lines if each of those functions call to a templated function in their
implementation called Compute (that's why this function is not
virtual). So I decided to give a try to the hierarchy generator (also
provided by loki) as follows:

template <class Class>
class ClassComputerUnit {

public:
virtual typename ClassVisitor::ReturnType Visit(Class& el) {
// call Compute function
}
virtual ~ClassComputerUnit() {}
};

class VisitorImpl : public ClassVisitor, public
GenScatterHierarchy<ClassList, ClassComputerUnit> {

// ... unrelated stuff

// create visitor as a functor
void operator(Class_base& b) { b.Accept(*this); }

// all 20 Visit functions don't show up anymore, they are supposed
to be inherited through the
// GenScatterHierarchy generator
};


Now, what GenScatterHierarchy does, is to derive from
ClassComputerUnit<Class> where Class is each of those classes in the
ClassList. So now I'm supposed to have all 20 Visit functions in the
VisitorImpl class overide the one of the Visitor. However, this is not
the case and I cannot compile my code because there are pure virtual
functions. So in a sense the GenScatterHierarchy did not overide the
Visit function from the visitor as I thought it would.

Does anyone have a clue why is this happening? I'm using GCC4.3.
Thanks in advance for the time reading such a long post.

aa
 
M

mlimber

Hello everyone,

I've been trying to work with the visitor design pattern, and it works
fine except for the following.
Let's suppose that we have a fixed hierarchy of classes (many of them)
which I cannot modify. I decided to use the visitor design pattern
depending on the actual type of the classes because those classes
already support the loki visitor. [...]
Now, what GenScatterHierarchy does, is to derive from
ClassComputerUnit<Class> where Class is each of those classes in the
ClassList. So now I'm supposed to have all 20 Visit functions in the
VisitorImpl class overide the one of the Visitor. However, this is not
the case and I cannot compile my code because there are pure virtual
functions. So in a sense the GenScatterHierarchy did not overide the
Visit function from the visitor as I thought it would.

Does anyone have a clue why is this happening? I'm using GCC4.3.
Thanks in advance for the time reading such a long post.

This is rather beyond the scope of this group, IMHO (cf.
http://www.parashift.com/c++-faq-lite/how-to-post.html#faq-5.9). Have
you asked on the Loki help forum on Sourceforge? Or can you boil it
down to a standard C++ question?

Cheers! --M
 
A

aaragon

Hello everyone,
I've been trying to work with the visitor design pattern, and it works
fine except for the following.
Let's suppose that we have a fixed hierarchy of classes (many of them)
which I cannot modify. I decided to use the visitor design pattern
depending on the actual type of the classes because those classes
already support the loki visitor. [...]
Now, what GenScatterHierarchy does, is to derive from
ClassComputerUnit<Class> where Class is each of those classes in the
ClassList. So now I'm supposed to have all 20 Visit functions in the
VisitorImpl class overide the one of the Visitor. However, this is not
the case and I cannot compile my code because there are pure virtual
functions. So in a sense the GenScatterHierarchy did not overide the
Visit function from the visitor as I thought it would.
Does anyone have a clue why is this happening? I'm using GCC4.3.
Thanks in advance for the time reading such a long post.

This is rather beyond the scope of this group, IMHO (cf.http://www.parashift.com/c++-faq-lite/how-to-post.html#faq-5.9). Have
you asked on the Loki help forum on Sourceforge? Or can you boil it
down to a standard C++ question?

Cheers! --M

It is a standard C++ question. It does not matter which kind of
visitor you use, all of them boil down to having a pure virtual
function in a base class, and have all subclasses override that
function. The GenScatterHiearchy class template is basically a
metaprogram, so it is a shorthand version of public Class1, public
Class2, and so on.

aa
 
M

mlimber

Hello everyone,
I've been trying to work with the visitor design pattern, and it works
fine except for the following.
Let's suppose that we have a fixed hierarchy of classes (many of them)
which I cannot modify. I decided to use the visitor design pattern
depending on the actual type of the classes because those classes
already support the loki visitor. [...]
Now, what GenScatterHierarchy does, is to derive from
ClassComputerUnit<Class> where Class is each of those classes in the
ClassList. So now I'm supposed to have all 20 Visit functions in the
VisitorImpl class overide the one of the Visitor. However, this is not
the case and I cannot compile my code because there are pure virtual
functions. So in a sense the GenScatterHierarchy did not overide the
Visit function from the visitor as I thought it would.
Does anyone have a clue why is this happening? I'm using GCC4.3.
Thanks in advance for the time reading such a long post.
This is rather beyond the scope of this group, IMHO (cf.http://www.parashift.com/c++-faq-lite/how-to-post.html#faq-5.9). Have
you asked on the Loki help forum on Sourceforge? Or can you boil it
down to a standard C++ question?
Cheers! --M

It is a standard C++ question. It does not matter which kind of
visitor you use, all of them boil down to having a pure virtual
function in a base class, and have all subclasses override that
function. The GenScatterHiearchy class template is basically a
metaprogram, so it is a shorthand version of public Class1, public
Class2, and so on.

MFC, Qt, and many other libraries would also qualify under such a
definition of standard C++. Can you rephrase the question so that we
can answer it by looking at the C++ standard? One way to do that is to
boil it down to a minimal but complete program that demonstrates your
problem. (Note, you'll likely get more help if it doesn't require
downloading Loki.)

Cheers! --M
 
A

aaragon

Hello everyone,
I've been trying to work with the visitor design pattern, and it works
fine except for the following.
Let's suppose that we have a fixed hierarchy of classes (many of them)
which I cannot modify. I decided to use the visitor design pattern
depending on the actual type of the classes because those classes
already support the loki visitor.
[...]
Now, what GenScatterHierarchy does, is to derive from
ClassComputerUnit<Class> where Class is each of those classes in the
ClassList. So now I'm supposed to have all 20 Visit functions in the
VisitorImpl class overide the one of the Visitor. However, this is not
the case and I cannot compile my code because there are pure virtual
functions. So in a sense the GenScatterHierarchy did not overide the
Visit function from the visitor as I thought it would.
Does anyone have a clue why is this happening? I'm using GCC4.3.
Thanks in advance for the time reading such a long post.
This is rather beyond the scope of this group, IMHO (cf.http://www.parashift.com/c++-faq-lite/how-to-post.html#faq-5.9). Have
you asked on the Loki help forum on Sourceforge? Or can you boil it
down to a standard C++ question?
Cheers! --M
It is a standard C++ question. It does not matter which kind of
visitor you use, all of them boil down to having a pure virtual
function in a base class, and have all subclasses override that
function. The GenScatterHiearchy class template is basically a
metaprogram, so it is a shorthand version of public Class1, public
Class2, and so on.

MFC, Qt, and many other libraries would also qualify under such a
definition of standard C++. Can you rephrase the question so that we
can answer it by looking at the C++ standard? One way to do that is to
boil it down to a minimal but complete program that demonstrates your
problem. (Note, you'll likely get more help if it doesn't require
downloading Loki.)

Cheers! --M

Ok, here we go, I thought that it was straightforward enough so it
would save me some coding. For simplicity let's assume that the return
type is void and that we only have three classes. Here it is:

#include <iostream>

using std::cout;
using std::endl;

class Base;
class Class1;
class Class2;
class Class3;

class ClassVisitor {
public:
void visit(Class1& ) {
cout<<"inside Class1"<<endl;
}
void visit(Class2& ) {
cout<<"inside Class2"<<endl;
}
void visit(Class3& ) {
cout<<"inside Class3"<<endl;
}
};


struct Base {

virtual void accept(ClassVisitor& vis) = 0;
virtual ~Base() {}
};


struct Class1 : public Base {
virtual void accept(ClassVisitor& vis) {
vis.visit(*this);
}
};
struct Class2 : public Base {
virtual void accept(ClassVisitor& vis) {
vis.visit(*this);
}
};
struct Class3 : public Base {
virtual void accept(ClassVisitor& vis) {
vis.visit(*this);
}
};


int main(int argc, char* argv[]) {


Class1 c1;
Class2 c2;
Class3 c3;

ClassVisitor vis;

c1.accept(vis);
c2.accept(vis);
c3.accept(vis);

return 0;
}


This prints:

inside Class1
inside Class2
inside Class3

Now, I want to change this in order to avoid having all those
declarations in the visitor class. So instead of having

class ClassVisitor {
public:
void visit(Class1& ) {
cout<<"inside Class1"<<endl;
}
void visit(Class2& ) {
cout<<"inside Class2"<<endl;
}
void visit(Class3& ) {
cout<<"inside Class3"<<endl;
}
};

I would like to have

template <class Class>
struct GenClass {
virtual void visit(Class& ) {
cout<<"inside class defined by template"<<endl;
}
};

class ClassVisitor :
public GenClass<Class1>, public GenClass<Class2>, public
GenClass<Class3> {
};

but when I try to compile this code I get ambiguous calls because of
course in the concrete classes, vis.visit(*this); this can be either
Class1, Class2 or Class3.

Now, I would like to solve this problem so I'm asking if anyone knows
how to do it. Thanks,

aa
 
A

aaragon

* aaragon:


Hello everyone,
I've been trying to work with the visitor design pattern, and it works
fine except for the following.
Let's suppose that we have a fixed hierarchy of classes (many of them)
which I cannot modify. I decided to use the visitor design pattern
depending on the actual type of the classes because those classes
already support the loki visitor.
[...]
Now, what GenScatterHierarchy does, is to derive from
ClassComputerUnit<Class> where Class is each of those classes in the
ClassList. So now I'm supposed to have all 20 Visit functions in the
VisitorImpl class overide the one of the Visitor. However, this is not
the case and I cannot compile my code because there are pure virtual
functions. So in a sense the GenScatterHierarchy did not overide the
Visit function from the visitor as I thought it would.
Does anyone have a clue why is this happening? I'm using GCC4.3.
Thanks in advance for the time reading such a long post.
This is rather beyond the scope of this group, IMHO (cf.http://www.parashift.com/c++-faq-lite/how-to-post.html#faq-5.9). Have
you asked on the Loki help forum on Sourceforge? Or can you boil it
down to a standard C++ question?
Cheers! --M
It is a standard C++ question. It does not matter which kind of
visitor you use, all of them boil down to having a pure virtual
function in a base class, and have all subclasses override that
function. The GenScatterHiearchy class template is basically a
metaprogram, so it is a shorthand version of public Class1, public
Class2, and so on.
MFC, Qt, and many other libraries would also qualify under such a
definition of standard C++. Can you rephrase the question so that we
can answer it by looking at the C++ standard? One way to do that is to
boil it down to a minimal but complete program that demonstrates your
problem. (Note, you'll likely get more help if it doesn't require
downloading Loki.)
Cheers! --M
Ok, here we go, I thought that it was straightforward enough so it
would save me some coding. For simplicity let's assume that the return
type is void and that we only have three classes. Here it is:
#include <iostream>
using std::cout;
using std::endl;
class Base;
class Class1;
class Class2;
class Class3;
class ClassVisitor {
public:
void visit(Class1& ) {
cout<<"inside Class1"<<endl;
}
void visit(Class2& ) {
cout<<"inside Class2"<<endl;
}
void visit(Class3& ) {
cout<<"inside Class3"<<endl;
}
};
struct Base {
virtual void accept(ClassVisitor& vis) = 0;
virtual ~Base() {}
};
struct Class1 : public Base {
virtual void accept(ClassVisitor& vis) {
vis.visit(*this);
}
};
struct Class2 : public Base {
virtual void accept(ClassVisitor& vis) {
vis.visit(*this);
}
};
struct Class3 : public Base {
virtual void accept(ClassVisitor& vis) {
vis.visit(*this);
}
};
int main(int argc, char* argv[]) {
Class1 c1;
Class2 c2;
Class3 c3;
ClassVisitor vis;

return 0;
}
This prints:
inside Class1
inside Class2
inside Class3
Now, I want to change this in order to avoid having all those
declarations in the visitor class. So instead of having
class ClassVisitor {
public:
void visit(Class1& ) {
cout<<"inside Class1"<<endl;
}
void visit(Class2& ) {
cout<<"inside Class2"<<endl;
}
void visit(Class3& ) {
cout<<"inside Class3"<<endl;
}
};
I would like to have
template <class Class>
struct GenClass {
virtual void visit(Class& ) {
cout<<"inside class defined by template"<<endl;
}
};
class ClassVisitor :
public GenClass<Class1>, public GenClass<Class2>, public
GenClass<Class3> {
};
but when I try to compile this code I get ambiguous calls because of
course in the concrete classes, vis.visit(*this); this can be either
Class1, Class2 or Class3.
Now, I would like to solve this problem so I'm asking if anyone knows
how to do it. Thanks,

Why don't you make ClassVisitor::visit a template, instead of making
ClassVisitor a template.

Since it's a callback I'd also prefer to change the name, e.g. to 'onVisitTo',
not 'visit' which implies an action that can be called in order to visit.

But that is of course at least to large degree a personal preference.

Cheers, & hth.,

- Alf

--
A: Because it messes up the order in which people normally read text.
Q: Why is it such a bad thing?
A: Top-posting.
Q: What is the most annoying thing on usenet and in e-mail?

Thanks for replying. I tried that in my example, and even though it
works for this simplified version, it doesn't work for the real thing.
I still have pure virtual function errors. I tried

template <class T>
ReturnType Visit(T&) {
cout<<"here!"<<endl;
exit(1);
}

inside the visitor but I still get

test.cpp:113: error: cannot declare variable 'a' to be of abstract
type 'yafeq::Test::assemble(yafeq::equation_type*) [with A =
yafeq::Assembler]::AssemblerType'
assembler.hpp:36: note: because the following virtual functions are
pure within 'yafeq::Test::assemble(fea::equation_type*) [with A =
yafeq::Assembler]::AssemblerType':
util.hpp:376: note: R cpputil::Visitor<T, R>::Visit(T&) [with T =
fea::Class<PROP1>, R = void]
util.hpp:376: note: R cpputil::Visitor<T, R>::Visit(T&) [with T =
fea::Class<PROP2>, R = void]
make[2]: *** [test.o] Error 1
make[1]: *** [all-recursive] Error 1
make: *** [all] Error 2


and I don't know why it's doing this. Any ideas?
 
M

mlimber

Hello everyone,
I've been trying to work with the visitor design pattern, and it works
fine except for the following.
Let's suppose that we have a fixed hierarchy of classes (many of them)
which I cannot modify. I decided to use the visitor design pattern
depending on the actual type of the classes because those classes
already support the loki visitor.
[...]
Now, what GenScatterHierarchy does, is to derive from
ClassComputerUnit<Class> where Class is each of those classes in the
ClassList. So now I'm supposed to have all 20 Visit functions in the
VisitorImpl class overide the one of the Visitor. However, this is not
the case and I cannot compile my code because there are pure virtual
functions. So in a sense the GenScatterHierarchy did not overide the
Visit function from the visitor as I thought it would.
Does anyone have a clue why is this happening? I'm using GCC4.3.
Thanks in advance for the time reading such a long post.
This is rather beyond the scope of this group, IMHO (cf.http://www.parashift.com/c++-faq-lite/how-to-post.html#faq-5.9). Have
you asked on the Loki help forum on Sourceforge? Or can you boil it
down to a standard C++ question?
Cheers! --M
It is a standard C++ question. It does not matter which kind of
visitor you use, all of them boil down to having a pure virtual
function in a base class, and have all subclasses override that
function. The GenScatterHiearchy class template is basically a
metaprogram, so it is a shorthand version of public Class1, public
Class2, and so on.
MFC, Qt, and many other libraries would also qualify under such a
definition of standard C++. Can you rephrase the question so that we
can answer it by looking at the C++ standard? One way to do that is to
boil it down to a minimal but complete program that demonstrates your
problem. (Note, you'll likely get more help if it doesn't require
downloading Loki.)
Cheers! --M

Ok, here we go, I thought that it was straightforward enough so it
would save me some coding. For simplicity let's assume that the return
type is void and that we only have three classes. Here it is:

#include <iostream>

using std::cout;
using std::endl;

class Base;
class Class1;
class Class2;
class Class3;

class ClassVisitor {
public:
        void visit(Class1& ) {
                cout<<"inside Class1"<<endl;
        }
        void visit(Class2& ) {
                cout<<"inside Class2"<<endl;
        }
        void visit(Class3& ) {
                cout<<"inside Class3"<<endl;
        }

};

struct Base {

        virtual void accept(ClassVisitor& vis) = 0;
        virtual ~Base() {}

};

struct Class1 : public Base {
        virtual void accept(ClassVisitor& vis)  {
                vis.visit(*this);
        }};

struct Class2 : public Base {
        virtual void accept(ClassVisitor& vis)  {
                vis.visit(*this);
        }};

struct Class3 : public Base {
        virtual void accept(ClassVisitor& vis)  {
                vis.visit(*this);
        }

};

int main(int argc, char* argv[]) {

        Class1 c1;
        Class2 c2;
        Class3 c3;

        ClassVisitor vis;

        c1.accept(vis);
        c2.accept(vis);
        c3.accept(vis);

        return 0;

}

This prints:

inside Class1
inside Class2
inside Class3

Now, I want to change this in order to avoid having all those
declarations in the visitor class. So instead of having

class ClassVisitor {
public:
        void visit(Class1& ) {
                cout<<"inside Class1"<<endl;
        }
        void visit(Class2& ) {
                cout<<"inside Class2"<<endl;
        }
        void visit(Class3& ) {
                cout<<"inside Class3"<<endl;
        }

};

I would like to have

template <class Class>
struct GenClass {
        virtual void visit(Class& ) {
                cout<<"inside class defined by template"<<endl;
        }

};

class ClassVisitor :
public GenClass<Class1>, public GenClass<Class2>, public
GenClass<Class3> {

};

but when I try to compile this code I get ambiguous calls because of
course in the concrete classes, vis.visit(*this); this can be either
Class1, Class2 or Class3.

Now, I would like to solve this problem so I'm asking if anyone knows
how to do it. Thanks,

Now I see the problem. It's the one described in this FAQ:

http://www.parashift.com/c++-faq-lite/templates.html#faq-35.19

You can either modify ClassVisitor:

struct ClassVisitor
: public GenClass<Class1>
, public GenClass<Class2>
, public GenClass<Class3>
{
using GenClass<Class1>::visit;
using GenClass<Class2>::visit;
using GenClass<Class3>::visit;
};

Or you can change the classes themselves:

struct Class1 : Base
{
virtual void accept( ClassVisitor& vis )
{
vis.GenClass<Class1>::visit( *this );
}
};

Neither of these are ideal, but perhaps Loki has solved (or could
solve) this problem with macros and type lists so that ClassVisitor
looks something like:

typedef TypeList<Class1, Class2, Class3> MyTypeList;

struct ClassVisitor
: GenHierarchy< MyTypeList >
{
GEN_USINGS( MyTypeList, visit );
};

Cheers! --M
 
A

aaragon

Hello everyone,
I've been trying to work with the visitor design pattern, and it works
fine except for the following.
Let's suppose that we have a fixed hierarchy of classes (many of them)
which I cannot modify. I decided to use the visitor design pattern
depending on the actual type of the classes because those classes
already support the loki visitor.
[...]
Now, what GenScatterHierarchy does, is to derive from
ClassComputerUnit<Class> where Class is each of those classes in the
ClassList. So now I'm supposed to have all 20 Visit functions in the
VisitorImpl class overide the one of the Visitor. However, this is not
the case and I cannot compile my code because there are pure virtual
functions. So in a sense the GenScatterHierarchy did not overide the
Visit function from the visitor as I thought it would.
Does anyone have a clue why is this happening? I'm using GCC4.3.
Thanks in advance for the time reading such a long post.
This is rather beyond the scope of this group, IMHO (cf.http://www.parashift.com/c++-faq-lite/how-to-post.html#faq-5.9). Have
you asked on the Loki help forum on Sourceforge? Or can you boil it
down to a standard C++ question?
Cheers! --M
It is a standard C++ question. It does not matter which kind of
visitor you use, all of them boil down to having a pure virtual
function in a base class, and have all subclasses override that
function. The GenScatterHiearchy class template is basically a
metaprogram, so it is a shorthand version of public Class1, public
Class2, and so on.
MFC, Qt, and many other libraries would also qualify under such a
definition of standard C++. Can you rephrase the question so that we
can answer it by looking at the C++ standard? One way to do that is to
boil it down to a minimal but complete program that demonstrates your
problem. (Note, you'll likely get more help if it doesn't require
downloading Loki.)
Cheers! --M
Ok, here we go, I thought that it was straightforward enough so it
would save me some coding. For simplicity let's assume that the return
type is void and that we only have three classes. Here it is:
#include <iostream>
using std::cout;
using std::endl;
class Base;
class Class1;
class Class2;
class Class3;
class ClassVisitor {
public:
void visit(Class1& ) {
cout<<"inside Class1"<<endl;
}
void visit(Class2& ) {
cout<<"inside Class2"<<endl;
}
void visit(Class3& ) {
cout<<"inside Class3"<<endl;
}

struct Base {
virtual void accept(ClassVisitor& vis) = 0;
virtual ~Base() {}

struct Class1 : public Base {
virtual void accept(ClassVisitor& vis) {
vis.visit(*this);
}};
struct Class2 : public Base {
virtual void accept(ClassVisitor& vis) {
vis.visit(*this);
}};
struct Class3 : public Base {
virtual void accept(ClassVisitor& vis) {
vis.visit(*this);
}

int main(int argc, char* argv[]) {
Class1 c1;
Class2 c2;
Class3 c3;
ClassVisitor vis;

return 0;

This prints:
inside Class1
inside Class2
inside Class3
Now, I want to change this in order to avoid having all those
declarations in the visitor class. So instead of having
class ClassVisitor {
public:
void visit(Class1& ) {
cout<<"inside Class1"<<endl;
}
void visit(Class2& ) {
cout<<"inside Class2"<<endl;
}
void visit(Class3& ) {
cout<<"inside Class3"<<endl;
}

I would like to have
template <class Class>
struct GenClass {
virtual void visit(Class& ) {
cout<<"inside class defined by template"<<endl;
}

class ClassVisitor :
public GenClass<Class1>, public GenClass<Class2>, public
GenClass<Class3> {

but when I try to compile this code I get ambiguous calls because of
course in the concrete classes, vis.visit(*this); this can be either
Class1, Class2 or Class3.
Now, I would like to solve this problem so I'm asking if anyone knows
how to do it. Thanks,

Now I see the problem. It's the one described in this FAQ:

http://www.parashift.com/c++-faq-lite/templates.html#faq-35.19

You can either modify ClassVisitor:

struct ClassVisitor
: public GenClass<Class1>
, public GenClass<Class2>
, public GenClass<Class3>
{
using GenClass<Class1>::visit;
using GenClass<Class2>::visit;
using GenClass<Class3>::visit;
};

Or you can change the classes themselves:

struct Class1 : Base
{
virtual void accept( ClassVisitor& vis )
{
vis.GenClass<Class1>::visit( *this );
}
};

Neither of these are ideal, but perhaps Loki has solved (or could
solve) this problem with macros and type lists so that ClassVisitor
looks something like:

typedef TypeList<Class1, Class2, Class3> MyTypeList;

struct ClassVisitor
: GenHierarchy< MyTypeList >
{
GEN_USINGS( MyTypeList, visit );
};

Cheers! --M

Thanks mlimber for replying. Indeed that is what is happening, and as
mentioned in the faq it is giving me a headache. I tried the "using"
way before posting and it didn't work. I cannot refer to those
functions defined in the base class with the "this" pointer because
the idea was to eliminate all code from the derived classes. Maybe it
is because of the way the visitor class in the loki library is defined
so I guess I have to come up with my own visitor class.

I'll keep trying and I will post my solution eventually. Thank you,

aa
 
M

mlimber

Hello everyone,
I've been trying to work with the visitor design pattern, and it works
fine except for the following.
Let's suppose that we have a fixed hierarchy of classes (many of them)
which I cannot modify. I decided to use the visitor design pattern
depending on the actual type of the classes because those classes
already support the loki visitor.
[...]
Now, what GenScatterHierarchy does, is to derive from
ClassComputerUnit<Class> where Class is each of those classes in the
ClassList. So now I'm supposed to have all 20 Visit functions in the
VisitorImpl class overide the one of the Visitor. However, this is not
the case and I cannot compile my code because there are pure virtual
functions. So in a sense the GenScatterHierarchy did not overide the
Visit function from the visitor as I thought it would.
Does anyone have a clue why is this happening? I'm using GCC4..3.
Thanks in advance for the time reading such a long post.
This is rather beyond the scope of this group, IMHO (cf.http://www.parashift.com/c++-faq-lite/how-to-post.html#faq-5.9). Have
you asked on the Loki help forum on Sourceforge? Or can you boil it
down to a standard C++ question?
Cheers! --M
It is a standard C++ question. It does not matter which kind of
visitor you use, all of them boil down to having a pure virtual
function in a base class, and have all subclasses override that
function. The GenScatterHiearchy class template is basically a
metaprogram, so it is a shorthand version of public Class1, public
Class2, and so on.
MFC, Qt, and many other libraries would also qualify under such a
definition of standard C++. Can you rephrase the question so that we
can answer it by looking at the C++ standard? One way to do that is to
boil it down to a minimal but complete program that demonstrates your
problem. (Note, you'll likely get more help if it doesn't require
downloading Loki.)
Cheers! --M
Ok, here we go, I thought that it was straightforward enough so it
would save me some coding. For simplicity let's assume that the return
type is void and that we only have three classes. Here it is:
#include <iostream>
using std::cout;
using std::endl;
class Base;
class Class1;
class Class2;
class Class3;
class ClassVisitor {
public:
        void visit(Class1& ) {
                cout<<"inside Class1"<<endl;
        }
        void visit(Class2& ) {
                cout<<"inside Class2"<<endl;
        }
        void visit(Class3& ) {
                cout<<"inside Class3"<<endl;
        }
};
struct Base {
        virtual void accept(ClassVisitor& vis) = 0;
        virtual ~Base() {}
};
struct Class1 : public Base {
        virtual void accept(ClassVisitor& vis)  {
                vis.visit(*this);
        }};
struct Class2 : public Base {
        virtual void accept(ClassVisitor& vis)  {
                vis.visit(*this);
        }};
struct Class3 : public Base {
        virtual void accept(ClassVisitor& vis)  {
                vis.visit(*this);
        }
};
int main(int argc, char* argv[]) {
        Class1 c1;
        Class2 c2;
        Class3 c3;
        ClassVisitor vis;
        c1.accept(vis);
        c2.accept(vis);
        c3.accept(vis);
        return 0;
}
This prints:
inside Class1
inside Class2
inside Class3
Now, I want to change this in order to avoid having all those
declarations in the visitor class. So instead of having
class ClassVisitor {
public:
        void visit(Class1& ) {
                cout<<"inside Class1"<<endl;
        }
        void visit(Class2& ) {
                cout<<"inside Class2"<<endl;
        }
        void visit(Class3& ) {
                cout<<"inside Class3"<<endl;
        }
};
I would like to have
template <class Class>
struct GenClass {
        virtual void visit(Class& ) {
                cout<<"inside class defined by template"<<endl;
        }
};
class ClassVisitor :
public GenClass<Class1>, public GenClass<Class2>, public
GenClass<Class3> {
};
but when I try to compile this code I get ambiguous calls because of
course in the concrete classes, vis.visit(*this); this can be either
Class1, Class2 or Class3.
Now, I would like to solve this problem so I'm asking if anyone knows
how to do it. Thanks,
Now I see the problem. It's the one described in this FAQ:

You can either modify ClassVisitor:
 struct ClassVisitor
    : public GenClass<Class1>
    , public GenClass<Class2>
    , public GenClass<Class3>
 {
    using GenClass<Class1>::visit;
    using GenClass<Class2>::visit;
    using GenClass<Class3>::visit;
 };
Or you can change the classes themselves:
 struct Class1 : Base
 {
   virtual void accept( ClassVisitor& vis )
   {
     vis.GenClass<Class1>::visit( *this );
   }
 };
Neither of these are ideal, but perhaps Loki has solved (or could
solve) this problem with macros and type lists so that ClassVisitor
looks something like:
 typedef TypeList<Class1, Class2, Class3> MyTypeList;
 struct ClassVisitor
    : GenHierarchy< MyTypeList >
 {
    GEN_USINGS( MyTypeList, visit );
 };
Cheers! --M

Thanks mlimber for replying. Indeed that is what is happening, and as
mentioned in the faq it is giving me a headache. I tried the "using"
way before posting and it didn't work. I cannot refer to those
functions defined in the base class with the "this" pointer because
the idea was to eliminate all code from the derived classes. Maybe it
is because of the way the visitor class in the loki library is defined
so I guess I have to come up with my own visitor class.

I'll keep trying and I will post my solution eventually. Thank you,

The two solutions I mentioned (using and explicit qualification)
worked for me just fine, and they should work for you. The using
solution is probably what you'll need to use.

Cheers! --M
 
A

aaragon

Hello everyone,
I've been trying to work with the visitor design pattern, and it works
fine except for the following.
Let's suppose that we have a fixed hierarchy of classes (many of them)
which I cannot modify. I decided to use the visitor design pattern
depending on the actual type of the classes because those classes
already support the loki visitor.
[...]
Now, what GenScatterHierarchy does, is to derive from
ClassComputerUnit<Class> where Class is each of those classes in the
ClassList. So now I'm supposed to have all 20 Visit functions in the
VisitorImpl class overide the one of the Visitor. However, this is not
the case and I cannot compile my code because there are pure virtual
functions. So in a sense the GenScatterHierarchy did not overide the
Visit function from the visitor as I thought it would.
Does anyone have a clue why is this happening? I'm using GCC4.3.
Thanks in advance for the time reading such a long post.
This is rather beyond the scope of this group, IMHO (cf.http://www.parashift.com/c++-faq-lite/how-to-post.html#faq-5.9). Have
you asked on the Loki help forum on Sourceforge? Or can you boil it
down to a standard C++ question?
Cheers! --M
It is a standard C++ question. It does not matter which kind of
visitor you use, all of them boil down to having a pure virtual
function in a base class, and have all subclasses override that
function. The GenScatterHiearchy class template is basically a
metaprogram, so it is a shorthand version of public Class1, public
Class2, and so on.
MFC, Qt, and many other libraries would also qualify under such a
definition of standard C++. Can you rephrase the question so that we
can answer it by looking at the C++ standard? One way to do that is to
boil it down to a minimal but complete program that demonstrates your
problem. (Note, you'll likely get more help if it doesn't require
downloading Loki.)
Cheers! --M
Ok, here we go, I thought that it was straightforward enough so it
would save me some coding. For simplicity let's assume that the return
type is void and that we only have three classes. Here it is:
#include <iostream>
using std::cout;
using std::endl;
class Base;
class Class1;
class Class2;
class Class3;
class ClassVisitor {
public:
void visit(Class1& ) {
cout<<"inside Class1"<<endl;
}
void visit(Class2& ) {
cout<<"inside Class2"<<endl;
}
void visit(Class3& ) {
cout<<"inside Class3"<<endl;
}
};
struct Base {
virtual void accept(ClassVisitor& vis) = 0;
virtual ~Base() {}
};
struct Class1 : public Base {
virtual void accept(ClassVisitor& vis) {
vis.visit(*this);
}};
struct Class2 : public Base {
virtual void accept(ClassVisitor& vis) {
vis.visit(*this);
}};
struct Class3 : public Base {
virtual void accept(ClassVisitor& vis) {
vis.visit(*this);
}
};
int main(int argc, char* argv[]) {
Class1 c1;
Class2 c2;
Class3 c3;
ClassVisitor vis;
c1.accept(vis);
c2.accept(vis);
c3.accept(vis);
return 0;
}
This prints:
inside Class1
inside Class2
inside Class3
Now, I want to change this in order to avoid having all those
declarations in the visitor class. So instead of having
class ClassVisitor {
public:
void visit(Class1& ) {
cout<<"inside Class1"<<endl;
}
void visit(Class2& ) {
cout<<"inside Class2"<<endl;
}
void visit(Class3& ) {
cout<<"inside Class3"<<endl;
}
};
I would like to have
template <class Class>
struct GenClass {
virtual void visit(Class& ) {
cout<<"inside class defined by template"<<endl;
}
};
class ClassVisitor :
public GenClass<Class1>, public GenClass<Class2>, public
GenClass<Class3> {
};
but when I try to compile this code I get ambiguous calls because of
course in the concrete classes, vis.visit(*this); this can be either
Class1, Class2 or Class3.
Now, I would like to solve this problem so I'm asking if anyone knows
how to do it. Thanks,
Now I see the problem. It's the one described in this FAQ:
http://www.parashift.com/c++-faq-lite/templates.html#faq-35.19
You can either modify ClassVisitor:
struct ClassVisitor
: public GenClass<Class1>
, public GenClass<Class2>
, public GenClass<Class3>
{
using GenClass<Class1>::visit;
using GenClass<Class2>::visit;
using GenClass<Class3>::visit;
};
Or you can change the classes themselves:
struct Class1 : Base
{
virtual void accept( ClassVisitor& vis )
{
vis.GenClass<Class1>::visit( *this );
}
};
Neither of these are ideal, but perhaps Loki has solved (or could
solve) this problem with macros and type lists so that ClassVisitor
looks something like:
typedef TypeList<Class1, Class2, Class3> MyTypeList;
struct ClassVisitor
: GenHierarchy< MyTypeList >
{
GEN_USINGS( MyTypeList, visit );
};
Cheers! --M
Thanks mlimber for replying. Indeed that is what is happening, and as
mentioned in the faq it is giving me a headache. I tried the "using"
way before posting and it didn't work. I cannot refer to those
functions defined in the base class with the "this" pointer because
the idea was to eliminate all code from the derived classes. Maybe it
is because of the way the visitor class in the loki library is defined
so I guess I have to come up with my own visitor class.
I'll keep trying and I will post my solution eventually. Thank you,

The two solutions I mentioned (using and explicit qualification)
worked for me just fine, and they should work for you. The using
solution is probably what you'll need to use.

Cheers! --M

Well, I mentioned that I was going to post my solution so here it is.
I took loki's cyclic visitor implementation and I changed so that I
can define visitors with non-strict visit behavior in a very nice
customizable way. So the solution was to implement a visitor that
falls back to another virtual function when the Visit(T&) is not
implemented instead of having the compiler complaint because of pure
virtual functions. This is called non-strict visit.


Here is the message that I put on the loki developer's forum:

Hello everyone,

I just got what I really wanted with the cyclic visitor. The cyclic
visitor class template takes now four template parameters: the return
type, the type list of objects to be visited, a mutable (immutable)
flag for non-const (const) objects and a template template parameter
indicating the default behavior when a function is not defined in the
visitor. In this way you can create a cyclic visitor as follows:

// defaults to mutable (non-const), do-nothing nonstrict visitor
typedef visitor::Visitor<ReturnType, ClassList> ClassVisitor1;

// do-nothing non-strict visitor for immutable objects (const)
typedef visitor::Visitor<ReturnType, ClassList, Immutable>
ClassVisitor2;

// personalize the behavior of the non-strict visitor
template <class R>
struct Functor1 {
template <class T>
R operator()(T&) {
// do whatever with T, when the function is not implemented
// in the visitor
}
};
typedef visitor::Visitor<ReturnType, ClassList, Immutable, Functor1>
ClassVisitor3;

// use strict visitor (compiler errors when R Visit(T&) not defined
// mutable
typedef visitor::Visitor<ReturnType, ClassList, Mutable, Strict>
ClassVisitor4;
// immutable
typedef visitor::Visitor<ReturnType, ClassList, Immutable, Strict>
ClassVisitor5;


I think this is much better than my previous post, even though I don't
think anyone has read that yet. The code is given below.

Best regards,

Alejandro M. Aragón

#ifndef VISITOR_HPP_
#define VISITOR_HPP_

#include <Loki/Typelist.h>

namespace visitor {

using Loki::NullType;
using Loki::Typelist;

// visitor class template, adapted from the Andrei Alexandrescu's
"Modern C++ Design"

enum Visit_type { Mutable, Immutable };

template <class T, typename R = void, Visit_type = Mutable>
class StrictVisitor;

template <class T, typename R>
class StrictVisitor<T, R, Mutable>
{
public:
typedef R ReturnType;
typedef T ParamType;
virtual ~StrictVisitor() {}
virtual ReturnType Visit(ParamType&) = 0;
};

template <class T, typename R>
class StrictVisitor<T, R, Immutable>
{
public:
typedef R ReturnType;
typedef const T ParamType;
virtual ~StrictVisitor() {}
virtual ReturnType Visit(ParamType&) = 0;
};

/// class template StrictVisitor (specialization)

template <class Head, class Tail, typename R>
class StrictVisitor<Typelist<Head, Tail>, R, Mutable>
: public StrictVisitor<Head, R, Mutable>, public StrictVisitor<Tail,
R, Mutable>
{
public:
typedef R ReturnType;
typedef Head ParamType;
// using StrictVisitor<Head, R>::Visit;
// using StrictVisitor<Tail, R>::Visit;
};

template <class Head, typename R>
class StrictVisitor<Typelist<Head, NullType>, R, Mutable> : public
StrictVisitor<Head, R, Mutable>
{
public:
typedef R ReturnType;
typedef Head ParamType;
using StrictVisitor<Head, R, Mutable>::Visit;
};

template <class Head, class Tail, typename R>
class StrictVisitor<Typelist<Head, Tail>, R, Immutable>
: public StrictVisitor<Head, R, Immutable>, public StrictVisitor<Tail,
R, Immutable>
{
public:
typedef R ReturnType;
typedef Head ParamType;
// using StrictVisitor<Head, R>::Visit;
// using StrictVisitor<Tail, R>::Visit;
};

template <class Head, typename R>
class StrictVisitor<Typelist<Head, NullType>, R, Immutable> : public
StrictVisitor<Head, R, Immutable>
{
public:
typedef R ReturnType;
typedef Head ParamType;
using StrictVisitor<Head, R, Immutable>::Visit;
};


////////////////////////////////////////////////////////////////////////////////
// class template NonStrictVisitor
// Implements non-strict visitation (you can implement only part of
the Visit
// functions)
////////////////////////////////////////////////////////////////////////////////

template <class R>
struct DefaultFunctor {
template <class T>
R operator()(T&) { return R(); }
};

template <class T, typename R = void, Visit_type V = Mutable, class F
= DefaultFunctor<R> > class BaseVisitorImpl;

template <class Head, class Tail, typename R, Visit_type V, class F>
class BaseVisitorImpl<Typelist<Head, Tail>, R, V, F>
: public StrictVisitor<Head, R, V>, public BaseVisitorImpl<Tail, R, V,
F> {
public:
typedef typename StrictVisitor<Head, R, V>::paramType ParamType;
virtual R Visit(ParamType& h)
{ return F()(h); }
};

template <class Head, typename R, Visit_type V, class F >
class BaseVisitorImpl<Typelist<Head, NullType>, R, V, F> : public
StrictVisitor<Head, R, V>
{
public:
typedef typename StrictVisitor<Head, R, V>::paramType ParamType;
virtual R Visit(ParamType& h)
{ return F()(h); }
};


/// Visitor

template <class R>
struct Strict {};

template <typename R, class TList, Visit_type V = Mutable, template
<class> class FunctorPolicy = DefaultFunctor>
class Visitor : public BaseVisitorImpl<TList, R, V, FunctorPolicy<R> >
{
public:
typedef R ReturnType;

template <class Visited>
ReturnType GenericVisit(Visited& host) {
StrictVisitor<Visited, ReturnType, V>& subObj = *this;
return subObj.Visit(host);
}
};


template <typename R, class TList, Visit_type V>
class Visitor<R, TList, V, Strict> : public StrictVisitor<TList, R, V>
{
public:
typedef R ReturnType;

template <class Visited>
ReturnType GenericVisit(Visited& host) {
StrictVisitor<Visited, ReturnType, V>& subObj = *this;
return subObj.Visit(host);
}
};

} // namespace visitor

#endif /* VISITOR_HPP_ */
 

Ask a Question

Want to reply to this thread or ask your own question?

You'll need to choose a username for the site, which only take a couple of moments. After that, you can post your question and our members will help you out.

Ask a Question

Members online

Forum statistics

Threads
473,968
Messages
2,570,149
Members
46,695
Latest member
StanleyDri

Latest Threads

Top